NOTE: Can the comments thread for this post be reserved for discussion of the matter at hand. The “main” discussion thread is here.
UPDATE: I have a piece on Newspoll and One Nation preferences in Crikey today. Unfortunately, my assertion that the only public statement on the matter had consisted of two tweets from Galaxy on Thursday was superseded by events by the time it was published – namely, that a report by Ben Packham of The Oz today reveals that Newspoll did, indeed, change its allocation of One Nation preferences in December (and not just before the most recent result, as per Barrie Cassidy of the ABC last week). However, the exact split of One Nation preferences being used by Newspoll remains a trade secret.
The subject of One Nation preferences has dominated opinion poll discussion ever since Newspoll came out with its surprise 51-49 result last week, so I thought I’d offer my own contribution by observing their behaviour at recent elections, known or surmised. To start with the former, we have hard data on how One Nation preferences divided between the Coalition and Labor at the five federal elections going back to 2004, and at the Western Australian state election last year:
|To L-NP||To ALP|
|WA 2017||17252||60.6%||11196||39.4%||29 out of 57|
|Federal 2016||88327||50.5%||86693||49.5%||15 out of 150|
|Federal 2013||12147||55.1%||9899||44.9%||15 out of 150|
|Federal 2010||14894||54.8%||12290||45.2%||21 out of 150|
|Federal 2007||17303||53.0%||15347||47.0%||35 out of 150|
|Federal 2004||78941||56.4%||61015||43.6%||77 out of 150|
The party was a weak presence between 2004 and 2013: it scored only 1.2% of the national vote when contesting over half the seats in 2004, then contested only a scattering of seats in 2007, 2010 and 2013. Nonetheless, its voters were highly consistent throughout this period in leaning modestly in favour of the Coalition. Then came the great One Nation revival in 2016 election, at which, sadly for pollsters and prognosticators, it only offered 15 candidates as a test of its preference behaviour. A seat by seat account goes as follows, with notes added for the two seats where they directed preferences to a particular party rather than running a split ticket, and the two where they were top of the ballot paper:
|To L-NP||To ALP|
|Blair (Qld)||6603||49.7%||6670||50.3%||HTV to L-NP|
|Longman (Qld)||3608||43.5%||4685||56.5%||HTV to ALP|
|Oxley (Qld)||3643||51.9%||3380||48.1%||Donkey vote to LNP|
|Paterson (NSW)||4735||36.3%||8321||63.7%||Donkey vote to ALP|
|Wide Bay (Qld)||7265||51.8%||6757||48.2%|
For whatever reason, the traditional Coalition lean disappeared on this occasion, with preferences splitting straight down the middle. Kevin Bonham points to the unusual circumstance of the party’s determination to oust Wyatt Roy in Longman, but excluding this result only increases the Coalition’s preference share from 50.5% to 50.8%. The bigger outlier is the extent of the flow to Labor in Paterson, which is presumably the effect of the donkey vote.
The reason a pollster might hesitate to accept this even split will be a permanent state of affairs is the failure of the last two state elections contested by One Nation to bear it out. As the top table shows, One Nation preferences clearly broke to the Liberals in Western Australia, no doubt reflecting the preference deal between the two, politically disastrous though it may have been. However, the result was even more pronounced in Queensland, albeit that we do not have hard data on preference flows to go on here.
What we can do though is try to model the effect of the One Nation vote on preference flows using the results of the thirty-eight seats that had One Nation candidates and Labor-versus-LNP final two-party counts. This is a particularly useful election for this exercise, as One Nation directed preferences in all seats on this occasion, and divided their recommendations fairly evenly between Labor and the LNP. It thus provides an incidental opportunity to measure the impact of their how-to-vote cards.
Against a dependent variable that records Labor’s share of preferences in each electorate, the model regresses a variable measuring One Nation’s share of the minor party vote (ONPshare); a variable with a value of 1 in seats where the One Nation how-to-vote card had Labor ahead of the LNP, and -1 for vice-versa (ONPdummy); and a variable that records Labor’s share of the major party vote, so as to record the natural tendency of preferences to favour the major party that is stronger in a given area (ALPvote):
Since all results are highly significant, we can surmise that the LNP would typically get 65.6% of One Nation preferences where the two major parties recorded the same primary vote and how-to-vote cards weren’t a factor (the intercept plus ONPshare plus half the value of ALPvote). Beyond that, the how-to-vote card could be expected to swing the total 4.6% in one direction or the other.
Given twelve of the fifteen seats contested by One Nation at the federal election were in Queensland, the very substantial difference between the two results is at least a little perplexing. However, by far the likeliest explanation is that One Nation’s gains since 2016 have been mostly Coalition supporters who are likely to feed their preferences back to their own party, if they do indeed end up voting for One Nation.
Adding further to the uncertainty is the tendency shown by One Nation, and minor parties more generally, to shed support during election campaigns – partly because of the shambolic spectacle they usually present during the campaign, and partly because they tend not to contest all the seats. To the extent that One Nation is punching above its weight in recording 7.5% in the BludgerTrack poll aggregate, there will be a corresponding understatement of real-world support for other parties – perhaps particularly in the case of the Coalition, who may end up recovering some of that “soft” One Nation support.